ChatGPT Work and the Family Court: Can AI Agents Help Litigants in Person Prepare Their Cases Safely?

What is ChatGPT Work?

On 9 July 2026, OpenAI introduced ChatGPT Work as an agent for longer, more involved tasks. Traditional conversational AI generally responds to individual questions. Work can go further: it can plan linked stages of a task, research sources, work across connected apps and files, and create finished documents, spreadsheets and reports.

Depending on account access, integrations and permissions, it can work across supported websites and assist with scheduled or recurring workflows. The significant development is a shift from AI that responds to a question towards AI that can take steps to complete a goal.

This does not make the system legally authoritative. It may still misunderstand source material, omit qualifications or produce convincing but inaccurate answers. The person using it remains responsible for checking the outcome.

Why does this matter for family proceedings?

Private children proceedings can involve years of information: court orders, Cafcass assessments, school records, medical information, communications, safeguarding allegations, professional reports, screenshots, witness statements and chronologies. A litigant in person may have many documents but limited help understanding what they establish.

AI could assist with blank evidence registers, organising dates, developing checklists and explaining publicly available guidance. Used carefully, it could help a person identify which information is missing and present a clearer, more focused document.

But the family court is not a document-production exercise. It must assess evidence, procedural fairness and, in children cases, welfare. An AI system cannot determine whether allegations are true, assess witness credibility reliably or decide what arrangements serve a child’s welfare.

Can you use ChatGPT in Family Court proceedings?

There is no general prohibition on litigants in person using AI to assist with preparing their own case. However, AI does not relax the obligation to give accurate information, comply with court directions or respect restrictions on sharing confidential documents.

In D (A Child) (Recusal) [2025] EWCA Civ 1570, decided on 9 December 2025, the Court of Appeal considered AI-assisted legal submissions which contained erroneous authorities. Baker LJ recognised in paragraph 83 why litigants in person may turn to AI, but stressed that it is not an infallible source of legal knowledge. Legal authorities must actually exist and support the propositions advanced. Read the official judgment.

The case does not establish that AI use is automatically improper. It establishes the practical importance of verification. A fictitious judgment or incorrectly described legal rule is not made acceptable by a persuasive-looking citation.

The judiciary’s October 2026 AI position statement

On 9 October 2026, the judiciary issued its Position Statement on the Judicial Use of Artificial Intelligence. It confirms a cautious but positive approach to adopting AI, with controlled testing of functions such as summarisation, transcription, redaction, accessible writing, chronology preparation and document analysis.

It is explicit that judicial decisions must be taken by judicial office holders. AI cannot replace their reasoning, evaluation of evidence or exercise of discretion. Every AI output used for judicial work must be reviewed, verified and approved by the responsible judicial office holder.

Approved AI tools for judicial purposes must have specified safeguards, including UK data residency, enterprise data protection and disabled web access. These requirements apply to tools authorised for judicial use; they are not automatically rules governing every litigant’s own software. Nevertheless, the statement underlines how much legal AI adoption depends on effective human oversight and security.

Five ways an AI agent could help a litigant in person

1. Explaining court procedure

AI can define unfamiliar terminology, identify relevant publicly available guidance and help organise questions about applications and hearings. Procedural deadlines and required forms must still be checked against the relevant order and authoritative rules.

2. Creating and checking chronologies

AI may help order dates, expose unexplained gaps and build a chronology template. The output should record sources and distinguish a disputed allegation from an observation, admission, professional opinion or judicial finding.

3. Reviewing professional reports

Where sharing and processing are lawful and suitably secure, AI may help locate which passages of a report rely on particular records or attributed accounts. It can support focused questions; it cannot establish that Cafcass or a social worker is wrong merely because material is disputed.

4. Preparing hearing materials

AI can generate blank structures for position statements, hearing notes, evidence schedules and document indexes. The litigant must check the content and remain responsible for approval, filing, service and compliance with directions.

5. Making legal information more accessible

AI can explain public guidance in plain English and assist with accessible formats. This may be useful for people who cannot afford comprehensive representation or who need help understanding unfamiliar processes. General information is not case-specific legal advice.

Confidentiality, children’s information and safeguarding

Family proceedings frequently include highly sensitive details about children and adults, including health records, protected addresses, school and local-authority information and allegations of domestic abuse. Uploading these materials to an AI provider is not automatically safe or lawful merely because an account is paid for or the system has commercial security features.

The Family Procedure Rules 2010, Part 12, including rules 12.73 to 12.75 and relevant Practice Directions, regulate communication of information in specified children proceedings. Whether sharing is permitted depends on the information, the purpose, any applicable exception, relevant orders and the recipient. Permission to share information confidentially with a properly authorised adviser does not automatically permit uploading it to an external AI platform.

UK GDPR and the Data Protection Act 2018 may also apply where processing falls within their scope. The Information Commissioner’s Office guidance on AI and data protection addresses relevant obligations and risks. For legal service providers, data retention, contracts, access controls, international transfers, and training or reuse of information require consideration.

What does the SRA say?

On 17 August 2026, the Solicitors Regulation Authority published a warning notice on misuse of AI. It highlights inaccurate generated authorities and the protection of confidential client information. SRA-regulated firms and individuals remain accountable for their own work, even if AI helped prepare it. The notice is directed at those regulated by the SRA; it does not impose those professional duties on every litigant in person.

A new problem: when AI agents can take action

An AI assistant that drafts a document can produce incorrect wording. An agent that also has access to connected applications may be able to misclassify records, change files, share information or act upon an unreliable external instruction. That creates a different level of operational risk.

The response should be practical: limit the agent’s permissions, preserve original evidence, use working copies, keep an audit trail and require human approval for consequential steps. Sensitive legal workflows should not be treated as fully automatic simply because the technology supports automation.

An agent should not independently file documents, contact the opposing party, amend original exhibits or send confidential correspondence without appropriate authority and review.

For more on this distinction, see JSH Law’s analysis of agentic AI, legal responsibility and confidentiality.

Testing an AI-generated document: JSH Law’s Six-Question Check

The JSH Law Six-Question Check is a practical case-preparation framework for examining an important conclusion or recommendation. It is not a legal test and does not substitute for the court’s evaluation of evidence or statutory welfare principles.

For example, an original report may say that a parent reported a safeguarding concern. An AI summary that omits the attribution and describes the concern as an established safeguarding incident has altered the evidential status. That is not a harmless stylistic edit.

See also JSH Law’s line-by-line guide to inaccurate social worker reports.

How can litigants in person use AI more safely?

1. Start with publicly available legal material or an empty template rather than private case files.
2. Check deadlines and requirements against the most recent sealed order and authoritative guidance.
3. Verify every legal citation, paragraph, quotation and factual statement against the original source.
4. Preserve original records and use separate working copies.
5. Keep disputed allegations, observations, professional assessments and findings distinct.
6. Assess confidentiality and legal authority before uploading sensitive information.
7. Review all generated material for proportionality, balance and impact on the child’s welfare.
8. Require human approval before external communication or any court filing or service.

An example of a safer instruction

Can ChatGPT Work replace a solicitor or McKenzie Friend?

No. AI can support general information and some document-preparation tasks, but it cannot take on the professional duties of an appropriately authorised adviser, exercise rights of audience or obtain authority to conduct litigation. The role of a McKenzie Friend is distinct and subject to court rules and judicial control.

Is an AI-generated summary evidence?

A generated summary does not authenticate its underlying records or prove that its statements are correct. The court evaluates the actual evidence under applicable rules and legal tests. A working summary should transparently identify its sources, qualifications and disputed material.

Better technology must mean better evidence, not more paperwork

The aim of legal AI should not be to produce longer submissions or more confident-looking arguments. A carefully checked chronology or concise position statement may help the court far more than a lengthy generated document with unclear sources.

The future of family justice should be technologically capable, evidence-led, safeguarding-aware and accountable. AI can improve access to information and reduce administrative burdens, but the responsibility to check the evidence, follow procedure and make judicial decisions must remain human.

Official primary sources

• OpenAI — ChatGPT Work announcement, 9 July 2026
• Judiciary — AI position statement, 9 October 2026
• Court of Appeal — D (A Child) (Recusal) [2025] EWCA Civ 1570
• SRA — Misuse of AI warning notice, 17 August 2026
• Family Procedure Rules 2010 — Part 12
• ICO — AI and data protection guidance

Related JSH Law guidance

• The JSH Law Six-Question Check
• How to Challenge an Inaccurate Social Worker Report
• AI agents, responsibility and confidentiality